[Dxspider-support] Hangs revisited

Dirk Koopman djk at tobit.co.uk
Sat Nov 19 18:20:55 GMT 2005


On Sat, 2005-11-19 at 09:22 -0700, Kelly Jones wrote:
> Has there been any resolution (or proposed fix) for the (seeming) inability 
> to take more than ~60 connections?  I currently have 36 users and 26 node 
> connections.  Spider simply refuses to allow any more.  My CPU load is 
> nearly nothing, but I am not able to log in any additional 
> nodes/users.  This is not a firewall/sessions/whatever problem.

But, as I said earlier: there are no limits in the code. Several people
on here (please speak up) have had connections running into the more
than 100. For instance: 

G1TLH-2 de GB7TLH 19-Nov-2005 1811Z dxspider > rcmd ed7zab-5 sh/cl
ED7ZAB-5: 172 nodes, 110 local / 1011 total users  Max users 1147
Uptime 7 01:34

A 'links' and count up of nodes adds another 26 or so. Which is already
more than double 60. 

I know that this is a pretty average number for Angel, he has had many
more connections.

> 
> It's as though the "listener" simply quits sending keyboard input back into 
> where it goes.  Connect establishes, you're greeted with the "enter your 
> call sign", then nothing.  Dirk, I know you have better things to do than 
> mess with the cluster, but I need to make some decisions based on whether 
> you've thought much about this.

I have thought about it and if you get a login prompt it would imply
that there is no problem with the actual tcpip connection, the handshake
must have happened. However, equally you can't be hitting some kind of
resource problem in your perl executable because the only one that is
likely is a lack of bits in the select vectors and if that were the case
you would not get the login prompt (which would already use a bit). 

I am frankly stumped and the only real thing I can suggest is to let me
in on ssh to observe and debug. 

Dirk G1TLH




More information about the Dxspider-support mailing list